A set of 500 unidentified users. At best one could say people within the group tend to do something (or don't do it). For example
Tezos.SE has 870 users (but let's call it 1000), so you could run a query on all of Tezos users (without userid) and receive an answer comprised from data using 50% of the users (randomly selected). Doing that 1000 times still wouldn't identify the specific action of a specific person.